documentary. $100,000 will be awarded in cash prizes. the grand prize of $5,000 will go to the student or team with the best overall entry. the deadline is january 18, 2018. so mark your calendar and help us spread the word to student film makers. for more information go to our website, student cam.org. snoop n . now the latest research on private retirement plans and social security benefits. it's concluded the number of seniors receiving private retirement benefits is on the rise. speakers include peter brady and u.s. census department representatives. >> thank you very much for coming today. i think we will get started.

welcome to all of the american enterprise institute. i'm mr. bigs, scholar here at aei, today's session is what can irs data tell us about retirement incomes. when i was give this task, i thought a b grade tag line of what if everything you knew were wrong. this is a modified version of the tag line. what if many of the things were wrong and what if many of those were about retirement incomes of it is a not as thrilling but it's something that is very important. and i think the fact is that much of what we think about we know about retirement incomes, the statistics that drive government policies regarding social security and retirement plans may be wrong. we think we know what retirement rates are, poverty rates are, how much they receive from private retirement plans.

can i pull some of those statistics from my head. the poverty rate for americans over 65 is a little bit under 10%. about a third of retiries receive almost all of their income from social security benefits. those are statistics that i knowis about on the data that's currently used. what we are going to find out today is some of those statistics are wrong because the data they are based on may be incorrect. in particular, the household survey data that's gathered by the government which asking people about their sources of income in retirement stmatically deflates incomes people receive from private plans. the result is we may come to skewed conclusions regarding how well off retirees are relative to other age groups. we may believe they are becoming dangerously depend oentan social security. that private plans failed and retirees are unable to maintain

the standard of living they enjoyed prior to retirement. the two studies presented today try to come to a more accurate view of retirement incomes by using more accurate data. in the process they tell a story that for a policy wonk at least is close to astonishing. based on government tax records the income for the typical retiree household is nearly one third higher than we previously thought. poverty is substantially lower and the typical retiree has a a income close to what he or she enjoyed in the years prior to retoormt. we don't know what the future will bring in light of increasing lifespans and the decades long shift from traditional defined benefit pensions to plans like 401(k)s and these papers don't make predictions regarding those future outcomes. more accurate data show americans retirement incomes

need to drop further before we could consider our retirement system nobody crisis. i'm happy to welcome our two presenters. first is peter brady, a senior economist in retirement in the investment company institute. following peter will be joshua mitchell a senior economist at the u.s. census bureau. to both of you thank you very much for coming and welcome to aei. both pete and josh are presenting group work. pete's study is coauthor is steven bass of ici and jessica holland and kevin pierce of the i.r.s. josh's paper is coauthored with adamby, also with the census bureau. we are also going to be two discussions on the work. kevin moore is chief of the economic survey session in the research and statistics division. and second produce myer, the mccormick foundation professor.

thank you both for coming today. following discussions comments, we will allow for a brief interchange between the participants and then we will turn to your questions. at this point i would like to begin with our presentations. pete, we have between 20 and 25 minutes. then we'll go to josh. thank you very much. >> thank you andrew, and thank you all for showing up today. as andrew already mentioned, this is a joint work. my coauthors are sitting in the front row. >> i also want to mention this paper is a product of the soi joint statistical research program. our paper examines two questions. first are workers able to maintain their standard of living after they claim social security benefits. to answer that what we are going to look at is changes in spendable income around the time they claim. secondly, where do retirees get their income from after they claim social security? to answer that we will look at the incidents and amount of income they get from different

sources. the reason we undertook this research is that the social security system and tax dematerial for retirement plans are important components of the u.s. tax and transfer system. together they generate most of the income that retirees rely on. and so it's very important for poliicy makers and administratos to understand how these programs are working. there are many proposals to change one or more of these programs and they are saying there is a real estate tirmt crisis, that is they are not saving enough while they are working to maintain their standard of living after they retire. what do we find. >> most are able to maintain their inflation adjusted net after they retire and the individuals who are the lowest income when we first observe them tend to replace the highest percentage of that income. where do they get their income? in addition to social security

it turns out most we observed in the data have income from pensions, annuities and iras, what we refer to as retirement income. indeed, most in the sample are going to rely on a mix of both social security benefits and this retirement income. the plan for today is i'm going to describe a little of the data. i'm going the look at how spendable income changes in retirement. then i'm going to -- after claiming social security. then i'm goc to look at where do individuals get their income from. then a brief discussion on if we can discern where the pension and annuity income, what is generating that income. okay, so the data we use is a panel data set. it is a 199 to 2010 panel. it starts with a representative sample of all tax payers in 1999. it's going to -- we are going to look at all individuals who are either a primary or secondary taxpayer on a joint return or primary taxpayer on a non-joint return. we are going to follow all those

individuals through 2010. what we want to do is observe them before they claim social security, observe them claiming and look at what happens after that claiming incident. so what we are going to focus on are people who are aged 55 to 61 in 1999. just below the early retirement claiming age who are working and have not received any social security benefits. then what we want to look at is what happens after they claim social security. one nice thing about focusing on this particular group is that the tax data appear to be representative of this group or study. so if you look at all individuals aged 55 to 61 and you compare the number in the tax returns and the number in population kouptsz, over 95% file a tax return. when you look at what we are looking at, which are those who are still working and haven't claimed social security the percentage is closer to 100%. we think most of the people who we want to look at are going to be in the tax data in 1999.

and they do not need to -- while they have to file a tax return in '99 you don't have to file a tax return in the later years to stay in the sample. the reason is we can grab wages and salary from the w-2, we grab social security benefits from the ssa 1099 and pension and annuity and ira distributions from the form 1 to 99r. we are going to -- we in -- the first year receiving social security income as the claiming year. and we refer to that as year t. and then we we are going the look at the five year period starting one year before you claim and extending out to three years after the year you claim. the -- when we rank individuals we are going to always rank them by their '99 income, which is when they are all working and before any claim social security. the income we may be measuring in the tables we are showing you

can be from as late as 2010 but we are always going to rank them by their '99 income when we show the income results. there is 12.5 million people in the tax data that meet our criterion of being 55 to 61 in '99, working and not having social security benefits. what we are going the analysis is the% that claim between 2000 and 2007. because we need to review their data three years after they claim. what excludes is 19% who we observed claiming in 2008, '09 '10. 5% who claimed social security before they claimed retirement benefits. 13% who were alive in 2010 and hadn't claimed social security benefits. and 4% who died before receiving the income. note for those who haven't received it or died before they received it we don't know if they were eligible to receive

the benefits. they may have been covered by a government plan or may not have had enough credits to qualify. i went through a lot of slides quickly. this summarizes all you need to know to go forward in the presentation. namely we are looking at people close to the early retirement age, early retirement claiming age that are working and don't have social security income. and then we are going follow anyone who claims between 20000 and 2007. we consider our first question. what happens to spendsible income. many studies of retirement will focus on how much of people's income or how much people's earnings they can replace in retirement. but economic theory suggests that people don't have a goal of replacing income or replacing earnings. in sort of english what they want to do is maintain their standard of living. at most what they would mean is they want to maintain how much spending they have in real terms, inflation adjusted terms.

although it's also possible they may be able to maintain their standard of living even if spending declines. so what we are going to look at, we can't measure spending in our data but we are going to look at spendable income. when we mean by that is simply what people have left over after they have saved and paid their taxes. the tax data do not allow us to measure, accurately measure savings as whole but we can directly measure retirement plan contributions and payroll taxes and federal income taxes. now, the results i'm going to show you today i'm going to focus on work-related income. what that means, includes is of course labor income, wages and salaries from work, also includes income from two other sources that ultimately are generated by working, that's social security benefits and distributions from pensions annuities and iras, it's going to exclude other taxable on taxable accounts, rents, royalties, business income, et cetera.

since we are focused on work-related income our measure of spendable income is going to be network related income. okay. so the next fewities slides what i'm going to show you are what happens to mean spendable income around the time of claiming. real that year t is the year in which we observe people claim social security benefits. it could be 2000 to 2007. then we look at years around that. the income measures are all inflation adjusts and per capita. which means for a married couple on joint return we split all the income equally. so what this slide shows that most people are able to maintain their spendable income per person after retirement. total work-related income declines by $4400 between t minus one and t plus 3 but taxes decline by $3600. net stays roughly constant over the period of time.

in the next few slides what we are going to look at is how this differs from individuals 1999 income. for knows in the lowest 20% of 1999 income they actually experienced on average a 29% increase in spendable income. for those in the middle quintile -- for those in the middle quintile income stays about flat and you don't start experiencing declines in income until you get to higher income individual. in this case, the 80th or 90th percentile. half are still working after claiming social security. the next slides is whether work status affects these results. what you see is a bigger increase for the lowest quintile. they experience a 43% increase in spendable income. for the middle quintile of workers who continue to work,

their income, net spendable income is about constant. and you see a decline at the top and it is a fairly modest decline. if we then look at what happens to those who aren't working three years later, the lower income actually still have an increase in spendable income, average spendable income but it's med modest 11%. in the middle, stays about conisn't. and spendable income at the top declines if you are higher income who no longer work. that shows us the mean on average gives you an idea of what is happening. of course there is a wide variation among people that the mean can mask. to look at how things vary and what the actual experiences of people are we calculate for every individual in the sample a replacement rate. the replatement rate we are looking at is a replacement of spendable income, not gross income. in particular we are focused on network related income. in all the replacement rates i'm going to show new the next few

slides is the denominator is always going to be spendable income in the year before they game, t-1. and then we will look at how that relates the next few slides will show the distribution of the results of everyone in the sample. the blue box is the median. everybody in the first year, the median replacement rate of spendable income is 106%, the green dots represent the 25th and 75th percent ail meaning 25% have a replacement of 88 or lower and 28% have a replacement of 126 or higher. and now if you go through t plus 3, the median stays constant although there's a slight increase in the spread of results. what i want to focus on again is year t plus 3 but i'm going to look at it by income. it's the same slide, we're going to do it by income. what you see is for the bottom 60% up through the middle of '99 income,

replacement rates on average, the median is over 100%. the other thing you see is that there's a general decline. '99 increases replacement rate and the whole distribution tends to shift down. again, about half of the individuals are working three years later so we want to see whether working and not working affects results. and what you can see is that for those who are working three years after claiming social security and what you can see is they have a higher replacement rate and the way you see that is that the entire distribution tends to shift up. you prepare the average, look at those working, it tends to shift up. but again you see the pattern where the highest replacement rates tend to be at the bottom and it slowly goes down. those who aren't working three years later tend to have lower replacement rates. and you can see that because the entire distribution of the replacement rates are shifting down. the general pattern is people with lower '99 income have

higher replacement rates and particularly lowest is over 100%. okay. so we find that most people on average are going to maintain their inflation adjusted spending and the lower income and those who were working three years later tend to have higher replacement rates. we can turn to the second question, which is where do people get their income from and for that we're going to look at incidents and amounts of income from different sources. what this slide illustrates is retirement is best thought of as a transitional period instead of a point in time. so what we see from this slide, if you look at the left, is even a year before they claimed social security, about 15% of those in the sample have stopped working. and the shared who are working continues to climb until three years later, although nearly half are working three years later. and 61% are working or have a spouse who is working.

a slow transitional period. what you also see as the amount of people working declines, the amount of people who have income from pensions, annuities and iras increase. so even the year before they claim social security, nearly half of individuals have income from these sources and that increases to 72% three years after they claim. and this income tends to be persistent in the 67% of those who claim in the first year, 92% of them claim all three years after claiming social security. so in the next few slides i'm going to focus on the three years after claiming and now i'm going to again look at it by '99 income and see what the pattern is. so what we see here is that for a large portion of the middle or upper parts, so from the 40th percentile to the 95th percentile, upper to the upper middle.

the loi lowest incidents we see is both in the top 5% and the lower 40%. if we extend this analysis out to the full five-year period and say how many people received the income in at least one of the five years, we get that 81% receive income from these sources. in addition, we have evidence that some of the people who never receive pension income actually have some of these resources. in particular, we can observe someone who had a 1099 r and didn't have pension income. when that happened, that's like a rollover to an ira. they got a distribution but didn't roll it over. didn't create any income. or we can see they have a form 5498, they have an ira and they're not yet drawing down income. so the upshot of this slide is that if you're 55 to 61 in '99 and you're not in the bottom of income, which means you have per capita income of $17,000, which

would be about $24,000 in today's term, you have a good chance of having these resources. nearly everybody at the top has these. okay. so that tells you they have them. how important it is. well to look at how important it is, what i want to look at is for everybody nonlabor income, i want to remove labor income because some are working and some aren't working. look at where the nonlabor income comes from and we're going to do it by the '99 income. so what this shows is that social security benefits are a much portion with lower income in '99, a much smaller portion at the high end. and this reflects the design of the benefit. this is the benefit formula. it's designed to replace a higher percent of earnings at the low end and higher earnings at the high end. income from pensions annuity and iras has the opposite pattern in that it's not as important at the low end but it tends to increase in importance going up. and what you actually see is for the people in the middle, the magnitude is on par with social

security benefits that they're getting. and then we don't really talk about the other income. but the wealthy are not like you and i, they have nonwork-related income of different types. okay. that shows on average where people are getting the income. of course, there can be lot of variation. of course we're going to show you and what we calculate for each individual we're going to calculate their income shares. and this is total income, including all sources of income and before taxes. what percentage comes from different sources. and what you see for the samplings as a whole, those who haven't claimed social security, the median amount they get is zero. there are some with social security benefits in there because they may have a spouse that had already claimed social security and that's included in the income. as they claim the percentage of income they get from social security increases and it's about median is about 43% but there's a wide variation. 10% of the sample only had social security benefits in year t plus 3.

if we look at retirement income, on we still see at the median there's very little retirement income the year before they claimed social security, although again about half of the people are getting this income before they claim. so there are some in the higher percentage from there. and as you go out to year t plus 3, the median is about 33%. so what that shows is for -- and then again we're going to look at the t plus 3 and see how -- this 43% and 33% bars, we're going to look at how that varies by income and what you see for the lowest 20% of '99 income, on average they get a lot of their income from social security. so the median is 77%, 25% get only social security benefits. okay. so the 75th percentile is at 100. very few are going to be getting retirement benefits and median amount of zero although there are some people with retirement benefits. if you go up the income distribution, you see the

pattern where social security becomes less important and retirement income becomes more important. for the middle income, again the median amount they get from social security, the median amount from retirement income are very similar. so -- i went one slide too many. okay. so again, these results were the most surprising results of studies. this is not what we expected when we went into the study. it is not what i thought i knew. i thought i knew where people were getting their income. and what we find is this income is very wide spread, nearly nine and ten have evidence of having resources from this. it's persistent and it tends to be a substantial on par for the middle income people with what they're getting from social security. although there is definitely a grading on this but that's reflected in the social security system. certainly the lower income are going to rely more on social security and the higher income are going to rely more on the retirement income.

those are the two questions that we set out to answer. however, our research has tended to prompt other to ask us questions that we are not necessarily well-inclined to answer but i'm going to take it on. and the question we often get is well this pension annuity income, where is it coming from? part of the interest is the shift from the db plans to dc plans in the private sector. so there's a big concern of how much is coming from private sector db plans. now it turns out that the tax data aren't well-designed to answer this question. from a tax perspectived perspective isn't really relevant. it's all taxed the same with the same tax treatment. so for example on the form 1040 what we know, we have two categories. one we know if it's a distribution and ira. in general we don't know whether it's pension or annuity or whether it's a db pension or a dc pension and we really don't know where the ira money is

they accept lump sum payments from db plans or dc plans or government plans or private sector plans. we really don't know the actual origin of that data. how much do we think is the maximum amount we say might be coming from the private sector pensions. what we're going to look at, the first thing we looked at was in our panel data set. we want to look at people who had pension income in two consecutive years and see how consistent that was.

our supposition is, well, if people tend to have about the same amount, so around 100% in bo both years, that might indicate a private sector pension. if they show a slight increase -- and so when we looked at it and looked at pension annuity income, what we found as far as percentage of individuals and percentage of dollars, about 1/3 falls into this category of near 100%. another 29% fall sbos ths into category of 101 to 105. this is private sector dbs and public sector dbs. what about people that receive

ira income in two years? what that show system that this 101 to 105 may be a good way to identify government people getting -- government dbs, but it turns out people have control over their accounts and have accounts and take regular payments sometime like getting the same amount each year. what we can say is about 1/3 of pension annuity income and about 1/3 of the retirement annuity income we're observing in t plus 3 is coming from this source. another way to look at it is to look at some aggregate data. this is from research we're in the process of writing up, analysis we've done using cross section data. this is 2010 data.

-- nearly $400 billion of that are transfers from one account to another. and that leaves about a little over $800 billion in net distributions. so the question is can we identify what percentage of that 827 is coming from private sector pensions or at least put a bound on it? so one way to do it is first start by looking at gross distributions and we know from the tax data that about 24% of this is iras and the rest is going to be pensions and annuities. then we can go to the aggregate data, or other sources of aggregate data and we can see about 61% of this is from what we identify what a db or dc pension, private sector db plans, governmental db plans. what's in the other 15%? well, we don't know. if we did, we'd tell you. but what we can tell by looking

at other data is that a substantial portion of this is going to be distributions from nonqualified annuities. these are reported on the same line. now, we can take some of the transfers out of specific items over there, so roth conversions are generally going come out of iras, and pensions but we don't know what type of pensions. wha what's the bottom line? we want to know what of that 827 could be coming from traditional db plans? we start with 127 coming out of private db plans. we can't use that full number because about $54 billion is coming from cash balance plans or what are often called high brit plans. only about $116 billion are coming from more traditional defined benefit plans. if all that income was taken out and none of it was lump sum distribution rolled through an ira, then we would get about 14%

of net distributions, measuring 2010 coming from the private sector pensions. this is something that unfortunately is very much of interest in the policy community. it's not important for filing taxes or determining your tax liabilities, so it's not something we can measure tax data, but we can try to put some bounds on it. okay. to summarize, our main findings is what we find is that most people are able to maintain their standard of living, their spendable income after they claim social security benefits and those who we observe in '99 that have the lowest income tend to have the highest replacement rates. we just find, again, a surprising amount of retirement income is widespread and it's persistent and it seems to be substantial. >> thank you very much, pete, and right on time. our next presenter this morning will be josh mitchell, the census bureau presenting joint

work he conducted. the views expressed today are mine and adams and not necessarily those of u.s. census bureau. >> i forgot to mention that, yes. >> so the motivation for this talk broadly is, the most important question in retirement policy, which is are americans saving enough for retirement, are they -- or are they in danger of running out of money at older ages. there are a variety of studies using different assumptions and methods that come to starkly different conclusions to this important question. one thing they nearly all have in common with the exception of pete's just released paper is they rely on one of a few popular household surveys to do their analysis. there is some long standing data quality concerns with income data in the household surveys. in particular, if you take a measure of income from a survey, aggregate it up, compare it to one of a few outside sources

such as the national income product accounts or tax data or other administration records which i have labeled as ad rec, you see the survey fuels short of the outside target. the key point is that by themselves these discrepancies can't tell us a lot. it could be on the one hand that there's a few wealthy households missing from the survey or who aren't reporting their income, in which case the statistics are largely unaffected. on the other hand, it could be that, if you found the missing income, put it back in the survey, it would broadly alter the understanding of living standards in this case for the older population. so the purpose of this paper is to answer that question. so we do that by linking our survey, census survey, at the microdata level so that we can compare the administrative records and the survey responses on a person by person basis. this will allow us to do distributional analysis. in particular, we'll develop new nationally representative estimates of median household income and poverty for the

population 65 and older through 2012. we're then going to step back and ask which of the administrative record data sources we use are most important for our findings. this will have implications for the importance of db versus dc income, the relative importance of social security income and we and we'll do a little bit of transition to retirement analysis as well. so the primary data source for this project is the u.s. census bureau's 2013 current population survey, annual social and economic supplement. this is in 2013 they asked about your income for the previous year. so 2012. and it's the source for the official 2012 income and poverty reports numbers. we have about 15,000 households with a household of 65 or older. in house we have a process that assigns a process identification key or pik to 90% of the sample which allows us to link from the

survey to other administrative record data sources. i'll focus on validating five types of income that are important for older americans. the first is earnings from wages and self employment. then social security benefit, ssi, dividends and interest and retirement income. which is apart from social security all the retirement income, survivor income, disability income. think of it broadly as db and dc income. the cps's long standing had a focus on money income concept which can be defined as a stream of regular payments. that fits very nicely with traditional defined benefit plans which typically paid an annuity upon retirement. there has been a shift from db plans to dc, and many retirees take withdrawal from dc plans on an irregular or as needed basis rather than regular basis. that's led many outside analysts to raise concerns that the cps was missing lots of dc income. there are also other types of

withdrawals. lump sums, rollovers, which poses a challenge to the cps, how do we think about measuring these. 2014 the cps underwent a major redesign to address these questions. they added specific questions about retirement account withdrawals and addressed some of the underreported asset income. my colleagues reevaluated and found a modest boost for the 2014 data. we don't have the administrative records for 2013, so everything we do today will be through the 2013 survey and we'll focus just on the traditional cps questions. in particular, the main cps retirement question asked, in the previous year did you or anyone in your household receive any pension or retirement income from a previous employer or any other retirement income apart from social security or va benefits which are questions asked earlier in the interview. if you say yes, we ask who received it, the amounts and the source.

if you could look at option number 7, you can see the language of regular payments from i.r.a.s, 401(k)s, et cetera. that's that language that led many outside analysts to raise concerns that most of the dc income would not be counted based on this definition. that's the survey. we linked the cps to several different administrative record sources three of which we obtained from the social security administration. in particular we get earnings information from w-2s and self-employment information and your monthly social security benefit amounts including deductions for medicare premiums as well as monthly ssi benefits, both federal and state supplements. from the irs we obtain two data sources. one, the form 1040. we have very limited fields from the 1040, so for today we're just using dividend income and tax and tax exempt amounts.

there is a complication with the 1040 because it's a return, not necessarily individual. so you have to make assumptions about who gets the asset income and what to do about non-filers. we split the asset income for joint filers and it's not going to matter much because we aggregate back up to the family or household level. for non-filers we continued to use the survey values for interest and dividends. lastly, we have this form 1099r which pete talked about, the gross distributions from pensions, annuities, retirement, profit-sharing plans, i.r.a.s, et cetera. you can think of this as capturing all the defined benefit income. and defined contribution withdrawals. it's an information return so you get one even if you do not file a 1040 in the given year. the file we obtain at census includes certain types of 1099r distributions and excludes others. in particular, we receive early distributions, disability and death distributions, but we exclude certain distributions

we would not wish to consider income including direct rollovers and conversions or recharacterizations which is more about moving money from one tax preferred account to another. we want to focus on measuring withdrawals when they leave the tax preferred universe for good. what we do is we take our survey and merge on the pik to all these different other datate da sources to measure income at the personal level. outside of these five sources that we're validating we use the cps values. we aggregate that back up to the family or household level as needed. as a first result we compare acrow -- aggregates between the cpa as a percent of the administrative record data source for two age groups, 18 to 64 and 65 and older. for earnings, the cps is capturing nearly 100% of earnings for both groups. for social security benefits it captures a high fraction.

particularly for those age 65 and older. for ssi things are a little bit more complex for younger workers. they are often reporting ssi on behalf of their children. when you look at the older population, which is the focus of this paper, we get about 70% of ssi benefits. when we get to interest and dividends, we're only capturing 60% for the older population. when you get to retirement income, we're only capturing 45%. when you put these five sources together for working age people we get about 95% of the total income because earnings dominate for that group. when you look at age 65 and older, we are missing about a quarter of the income for that group. so that's the aggregates. the key point of the paper is we can look at the distributional implications of the missing income. in particular we'll produce statistics similar to those produced in the annual income and poverty report, median household income for 65 and older and poverty rates persons 65 and older. poverty being defined as family

income compared to the relevant threshold which varies based on family size. i'll show results for three groups, the full cps sample, which replicates the official numbers, the subsets who are picked and the sample. kwh is which is the came as the pik sample we swap out the survey and replace them with the administrative records. the total was $33,800. looking at the sub set with piks, essentially the same. when we replaced it with the administrative records and recalculate it was $44,000. about a 30% difference. so quite substantial. i don't have time to go into all the results by demographic subgroups, but i will quickly flip through and demonstrate that there are large differences among most of the demographic subgroups examined. if we look at household 75 through 84, there is a large difference. for the older householders we

see a large difference in percentage terms. if you look at family householders. there are large differences, non-family householders, large differences. college graduate, large differences. those with some colleges, high schoolers, large differences. there is something there for those with less than a high school degree. we see it for white householders. black householders. asian, hispanic. so very pervasive across the socioeconomic distribution. we also did comparisons to outside data sources in particular the fed bulletin. which is based on the survey of consumer finances and that survey has a wealth of detailed information and a more expansive definition of income but counts the irregular withdrawals. it generally shows higher incomes than the cps. we do the comparison and still find that the administrative records are about 15% higher than the scf for those 65 and

older and also for 75 and older. that's the middle of the distribution. can these results affect the poverty rate? in 2012 the poverty rate was 9.1% for 65 and older. for the pirks pik sample it's 9%. when we replaced with the administrative records and recalculated, it falls over two points, to 6.9%. the findings are pervasive across age subgroups. you see a large difference for 65 to 74. 75 to 84 and 85 and older. one interesting thing, from the survey you see a high age gradient of poverty rate of four points. in the administrative records the total gap between the youngest and the oldest is only about one point. so that analysis replaced all the administrative records at once. a natural follow-up question is which of the administrative records data sources is driving our results, so we examine it

across the household income distribution, replacing one of the data sources at a time in a cumulative fashion and recalculate. this shows the baseline when i first swap out the survey earnings and replaced with the administrative records. we see that the red is essentially on top of the blue, earnings are well corresponding between the two data sources. next we swap out social security income. throughout most of the distribution things look good. there is some evidence of over reporting of social security at the bottom. turns out to be confusion among survey respondents between social security and the ssi program. when we also swap out ssi the over reporting goes away and things look pretty good throughout the distribution. next we swap out interest and dividends. we start to see a greater di vergens, particularly in the middle, and lastly when replacing retirement income we see the dramatic picture getting into 33% of the median and large differences throughout most of the household income distribution.

i want to pause for a minute and discuss some potential issues to be raised about the results. everything we've done as taken the records as the truth and as compared to the survey you could raise questions about the quality of the administrative records. we do a lot in the paper to validate our data against outside calculations. in general we think the tax data are of high quality and there isn't an incentive to over-report their income to the irs because it would trigger additional tax liability. with respect to another issue you could raise the quality of the linkage between the survey and the administrative records. we use the same linkage process. for all the administrative records. we're finding close correspondence for earnings and social security benefits so we don't think the data linkage is an issue. the last issue that could be raised is you can acknowledge that these resources are there but some do not wish to count these or view these as measures of pure income. that's a fair point, but we note

that in a world where we are transitioning from defined benefit to defined contribution, if you count the defined benefit income but not the dc, you'll perhaps get a distorted picture and it won't distinguish between two individuals, one who solely has social security income and nothing else and one who has social security income and makes substantial and regular withdrawals from the dc. do we qualify them as having the same living standard. i leave that question to you. next we examined the nature of under reporting in the survey and administrative records for the different income sources. and first, at the intensive margin we looked at people who both reported something in the survey and who had it in the administrative records. while earnings has the highest correlation we see fairly strong correlations for all the data sources. in particular retirement income is not badly reported relative to say, social security income. if you report retirement income, things look pretty good.

on the other hand, if you look at whether people say they have any of that type of income, for social security there is a close correspondence of 85% in both data sources. but when we look instead at retirement income only 37% say they have any in the cps whereas 61% have it according to administrative records. if you break down the 61%, it's about 50% have employer sponsored distributions and 29% have i.r.a.s. 61% having either of those. next we want to -- we do a lot of work in the paper asking what explains this retirement income under reporting. we threw every demographic variable we could at the problem but could only explain 5% of the variation in false negative rates. we looked at a variety of survey design features that could cause this such as whether you are the respondent, whether your observation was imputed. what month in the sample you were. we explained very little of the

variation based on these. we linked the retirement income 1099r to the community survey which is a different animal. where there is no interviewer. it's a mandatory survey. we get a similar rate of under reporting in the acs as to the cps. it's hard to pin it down on something specific to the cps. we get the most traction in terms of the nature of the 1099r income itself. in particular, if the false negative rates are very high for those who only have i.r.a. distributions but they're fairly for high for employer sponsored distribution. we also show that if you have a large amount of retirement income, you're more likely to report something in a cps, but with large amounts there's a fair amount of reporting. those who have more volatile distributions between one year and the next are less likely to report something in the survey. another issue that could be raised it whether the retirement income is a one-time event, in which case we might be distorting the picture of living standards. are these really regular or recurring payments. we looked at those who received

retirement income in 2012 and asked what fraction of them also received a distribution in the next year and what was the degree ever fluctuation between the two years, both for employer sponsored and i.r.a.s. for young people, it's actually a one-off event. it's not that common to have a distribution in consecutive years. when you get to 65 and older it's close to 100% have an employer sponsored distribution, 2013, if they had one in 2012. when we look at the fluctuation in amounts, those with a drop created the 10%. those with a small change, 10%. those with an increase, greater than 10%. at 65 and older people either have a steady change or, if they have a larger change, they're just as likely to have an increase or decrease. overall a very high fraction have either a small change or an increase. we look at i.r.a.s, again, for 65 and older you are likely to

have an i.r.a. distribution in 2013 if you had one in 2012. in part because of the rules which mandate you take distributions after age 70 1/2. we do look at the amounts, there is considerably more volatility with i.r.a. distributions. for example, for 75 to 84 only about half have a small change in i.r.a. amount. but we can see that, if anything, when they have a large change, they are more likely to have an increase in the next year than a large decrease. so, given that our results from the beginning of the presentation were so pervasive across the socioeconomic distribution and there even for the oldest retirees, we wanted to raise the question of whether the under-reporting was not simply the i.r.a.s and other dc income but whether defined benefit income was also missing from the cps. we did the analysis by looking at those in the cps linked to a 1099r from one of a few specific government retirement plans, either traditional db plans, so by anyone's definition this should be counted as money income.

and in particular we'll be looking at those who have ones from the federal civil service retirement system, military retirement fund or one of the larger state db plans. you can see that for those 65 and older in terms of having any employer sponsored distribution, about 35% say i have nothing in the cps. things are better for the government plans, but still a fraction of people are reporting zero retirement income in the cps and 25% for the federal civil service. 20% for the military retirement fund and 15% for calpers say they have zero. even though they are receiving the 1099rs. we wanted to examine how the boost in retirement income changes our understanding of the relative importance of different income sources across the income distribution. we used as our benchmark a publication by the social security administration, the income and age chart book based on the current population survey data. they use as their unit of

analysis an aged unit defined as an unmarried person 65 and older or married couple with husband 65 and older. we'll ask how the boost of retirement income affects all income sources. if i take the age units and rank them by administrative record income and group them, we can see the social security receipt rate correspond closely. in terms of amounts, again, we get pretty similar amounts across most of the distribution. in contrast, when we look at retirement income, for example, in the fifth, in the cps, 48% of aged units have some income. in administrative records it's 82%. quite a big difference. when we rate the administrative record amount into sub components, we do an analysis like pete did based on whether we think your distribution is defined benefit, employer dc or i.r.a., if we look at the fifth,

we see about 65% from some db income, about 35% have some i.r.a. income. only a small fraction have direct distribution from an employer dc account. most of that would have been rolled over, and we would see it when it comes out as an i.r.a. distribution. looking at amount of retirement income, of course, they're skewed. you can see large amounts. even in the fifth you can see the average amount is about $10,000, of which about $7500 we estimate is defined benefit income. this graph is done purely based on surveys. you can do this at home if you would like. we are showing the shares of income across the income distribution, and the big orange blob is the social security share, dark blue, the retirement income share. in the bottom decile it's 69% social security and the average share in the fifth is 72%. when we do this with the

administrative records we see that the share falls down to 55% and in the fifth to 53%. in the bottom it's misclassification between social security and ssi, the gray area got bigger. the fifth decile, the difference. in the bottom middle distribution this is not a misclassification of dollars it's a 30% higher income in the middle of the distribution. at the top, social security shares stayed about the same. the boost in the retirement income share comes at the expense of the earnings share. when we break down the type of retirement income across the distribution, retirement income overall is about 30% in the middle of the distribution and we get that 22% of the overall income is db. about 3/4 of the retirement income in the middle distribution is db as of 2012. the change in shares will affect

some statistics on the percent of social security beneficiaries where social security is half the income. if you use the cps, you get two-thirds rely on supreme court for 50% of income. with the administrative records it's half. in the 90% threshold we see the fraction that relies on social security for 90% of the incomes falls in half from 36% to 18%. everything i showed you so far was using income data from 2012. we linked the cps to many years of archives and records and extended the analysis back to a quarter century to look at trends in median incomes and poverty rates. in 1990 there is a discrepancy of 20% between the cps and administrative records. it grows to 30% in 2012. that means that, while median incomes grew 18% according to the survey, they grew 29% if you

used the linked data. in terms of poverty rates, the trends are largely unaffected by this. in every year you have a downward shift of over two points between the cps and the linked data. and one final graph showing the trends over time in retirement income received. we can see that at the beginning of the series the two are only a little bit apart between the cps and the linked data. and they diverge with the 61/36 difference in 2012. remarkably, according to the cps, there has been no increase at all in retirement income received and a big increase over time in the administrative records. lastly, i'll move quickly through this. we examine the retirement transitions similar to what pete has done and asked the question both in terms of a relative metric, how well the pre-retirement living standards were maintained as well as an absolute metric, do we observe increases in poverty. unlike pete's analysis we don't have a panel.

we're using a synthetic approach, which means we pull many of the linked samples together and examine similar individuals across years. the reason we do that is because that's the only way to compare the survey measure and administrative record measure across time. so these are just some details of the analysis. we looked at those claiming social security between 2003 and 2007. what we can see here is we do the results separately for men and women. years surrounding the claim of social security. according to cps you see the decline in income. median income in blue. we look at the same thing for administrative records. prior to claiming social security the two lines are on top of each other. after claiming when retirement in this case becomes more prevalent for men, you see the divergence here and much less of evidence of decline post-claiming social security when using the administrative records. we looked for women. we see less evidence of decline. there is a divergence that happens earlier because in fact

they often receive retirement income prior to claiming social security. looking at absolute measures such as poverty rates, we don't see much evidence of poverty rising in either series with lower poverty rates in the administrative records. looking at higher thresholds such as 200% of poverty or 300% of poverty. things look similar to men prior to claiming social security but there is a larger divergence after most ly due to retirement income. for women it happens earlier for the same reason as previously. to summarize that. incomes fall much more in the survey than in the administrative records. when there are income declines, they start several years prior to claiming, there isn't an abrupt change at retirement. we don't see evidence for men or women that poverty rates rise in either series. in most cases when there is a rise to 200% or 300% of poverty the rise is smaller in the administrative records than in the survey.

we think it's inconsistent with the notion that there is sudden hardship following retirement. to conclude with caveats. everything we did today only examines incomes of the population 65 and older through 2012. there will be changes as we shift from db to dc. we cannot easily extrapolate our findings. income may not capture all relevant aspects of material well-being. there is outside evidence that debt levels are rising among older households, if you have to use the higher income to service the debt rather than boost consumption, then maybe we're overstating things somewhat. finally, we examined the retirement transitions for abrupt changes around social security claiming. we can't use our analysis to ask the question what happens 20 years after retirement, et cetera. with those caveats, i will conclude and say thank you. here is our contacts. pass it on. >> thank you very much, josh.

that was extremely interesting. our first discussion today will be kevin moore of the federal reserve board. >> thanks, andrew. thanks for organizing this. interesting session. before i get started so i don't repeat or forget, i'll say the analysis and conclusion are mine and not those of the federal reserve board, the staff or anybody else at the federal reserve bank. so i am going to give brief comments on the two papers today. off the top, there are super interesting. as a survey practitioner and someone who runs the survey and finances. we're always interested to see if there are issues collecting data from people, especially josh and adam's paper points out that there could be issues about people reporting retirement income. also, as someone who uses tax data it's interesting to see how pete and his colleagues have really pushed the envelope on

using the information in the returns to bring good things to bear on the analysis. i'll start in reverse order here and go with adam and josh's paper first. overall from the two papers the headlines are, like andrew was saying, it looks like retirement income is higher than many people thought, from all the things we hear in the popular press or academic research. that has looked at this for a long time. some of it is due to things like the data sources may have been not as comprehensive as we thought. maybe there is no retirement consumption puzzle. it's because of this administrative data which has been sort of something that has come on the scenes over the last 15 or 20 years and really been used in a lot of studies these days. has to give the hat off to irs with their program to bring researchers in from the outside to be able to use this data and working with the people at soi who are super high quality. we've worked with them on the scf for 30 years.

it's been a great relationship. it looks like surveys are missing a substantial amount of retirement income. especially cps. what does this mean? well, it's good news. measures poverties are lower and replacement rates are lower. good news around. hard to see what would be bad about that. specifics about the bee and mitchell paper. they point out here when they're looking at administrative data versus cps, two headline things out of that is basically the 30% difference in income. i threw in a number that was similar to the one they gave for the 65 and older. it's for the -- 2012. you see our median is a little bit lower. that's a concern. i say that -- to defend our survey, if you look at the mean, there is a little bit more correspondence with the mean than the aggregate of retirement income. it matches up fairly well. there still may be under-reporting at the median that we should focus on.

i have a little more to say about that in a second. what's driving this all? this is the part, as a survey pr practitioner is super interesting to me. this chart from the paper -- if you had to look at one chart that explains the whole thing in the sense that retirement income. everything else when comparing the data, looks pretty good. retirement income, across the distribution here of household income is just, from basically about the fifth percentile out to 95% is 15, 30% higher. that's a pretty heavy rate of under reporting. the interesting thing about that is what's missing. it's not social security. people are getting social security, they seem to be able to know that and report it well in this survey and matches the administrative data. but it's the dc and db retirement plans. one of the things we've seen over time on the scf is, i guess, you know -- or just on any household survey or any survey that looks at retirement.

retirement is complicated for people. the amount of knowledge people have on it varies drastically. there is some obviously gradient with age. people who are younger aren't really thinking about retirement. even those who you think are on the doorstep of it seem sort of ill-informed about the choices and resources they have. here we have a case where people are supposedly in it and receiving income from it, and are still ill-informed. the interesting thing is that a lot of it is an the expensive margin. for these retirement plans. it's basically not that people are not sure how much they're getting. maybe they're reporting something after taxes or thinking about an amount -- they only keep half and give the spouse the other half or something along those lines. it's not that. it's that they're not even reporting that they have it. it's not just an issue for dc, for i.r.a.s, 401(k)s, you might have variable withdrawals. you don't get a check every month from your former employer.

you can understand that somewhat. it's also an issue for db plans which is pointed out. that was one of the most surprising thing for me. i would think the db plans would operate like social security. if you get a monthly payment you would remember that and know. as a survey practitioner, the variability in the dc income, it makes it hard to report. because these are things that people may not consider it a retirement income. they think of it like taking money out of my savings account. i took money out of my ira this month because i had some extra expense or just because i take $1,000 out of there every month or something. we thought long and hard in our survey about how to better capture that. so we ask about withdrawals from i.r.a.s after we ask about assets. right with the asset, we ask how much they have and then ask about withdrawals. we don't wait until we ask about income. at that point we feel like it's too distant from when we asked about the asset. we also do that with other types of dc plans. i think that helps us capture a bit more but still not perfect.

i guess if i had any silver lines is, again, for the people that reported they would be getting it fairly well. the problem is getting people to report it. it's also the case that on the aggregate, as your reporting improves with higher levels of income. it seems like that's good for capturing the aggregate but doesn't necessarily help with median households or households further down the distribution. moving on to pete's paper, like i said, or like he spoke of, they used the edited panel, one of the great products that soi produces and people use to research, really, really nice data. they look at individuals aged 55 to 61. they have data through 2010. this again is a period around claiming social security. as i mentioned before, one of the things that impressed me about this paper is they went all out on the information returns. i know that probably caused a lot of pain in terms of linking them together and bringing them all to bear.

i am sure that everything didn't just match up perfect the first time that they ran the job. even though the tax data is really good, there is still will always be issues. it's an impressive use of it. it really points to sort of innovations in using administrative data just beyond the 1040. i think that's something they're doing great work on. this is also a variety of different measures of income. the concept of work-related income. one thing is sort of as a tax policy person i was happy to see in the paper the account for this tax, the shift in taxation. you think about in general, it should be obvious that people, if they're not working and they move into retirement. payroll taxes go away. that's a big, big chunk of taxes for a lot of people. then distributional analysis also. one of the things you see from this is that social security is actually doing its job with the high replacement rates lower in the distribution.

the take away is most individuals are able to maintain their level of income. the takeaway from there would be one of the box plots showing replacement rates. it's nice to see how it goes out over the income distribution and how the replacement rates vary even across the top 1%. this is showing around this, or this group of people, around that time when they're climbing social security, it looks like things are looking fairly well in terms of replacement rates. so, i have a little bit less to say about pete's paper. i wasn't using the survey data. i think some of the issues for both papers, the main one to keep in mind is that this is a great result for the people who are in the cohorts that they're examining. they both -- both the authors are up front about this and not claiming what they're finding in their papers are sort of predictions for the future. they both talked about how the -- those cohorts have a pretty high incidence of db and

dc. some would argue they're still in the golden age of coverage for they have both types of plans. that is changing fast. we see it in our data coverage for younger cohorts. it's been declining and it's dc plans. it's not just going to work every day, putting in your time and getting the payout from whatever the formula is. it's actually there is going to be active saving that has to happen and there is variation in how people are saving. even two people, working at the same job for 20 years they could have vastly different savings amounts in their 401k at the end of that time due to saving behavior. these may not be what the future looks like in terms of whenever the people who are younger today get to that time period. one of the things -- one of the last things i'll touch on, because of the shift in dc plans is so important they've touched on that too, one of the things we're focused on in the scf is household wealth. one the things we're trying to capture better is retirement

assets. moving forward this pile of assets that people have will be more important than the stream of annuity payments they're getting from some db plan. a little takeaway from some of our ongoing work on measuring retirement wealth. we're trying to make an expand version. we ask a lot of questions about sort of wealth. one of the things we missed is we don't ask people to tell us the net present value of their social security benefits or defined benefit pension plans because nobody would know that answer. echb -- most economists would have a tough time. or it would devolve into a long discussion about mortality rates and everything else you should use and the interview would be six hours. but, the word -- we're using other data sources. amongst -- beside the scf and cps and financial accounts of the united states to allocate social security wealth to households and defined benefit pension wealth. this builds on work that's been done by many others.

also on it is hrs which had focused more on older households. we wanted to focus on younger. we just want to know where and how households are amassing retirement resources. i'll show you one chart from our work. this shows you for two age cohorts, 40-49 and 50 to 59 from three years from our survey, from '89 to 2013. the nonretirement wealth is everything you think of not including retirement accounts. and you can see how -- where the main wealth is for this group. then once you add in sort of dc wealth and our estimate of db pensions, then you see it has a pretty decent effect on mean wealth. the kick comes when you add the net present values of social security wealth. you do it for older households. it's what you expect. shift is even greater. the point is that basically we know we're not getting everything whenever we ask people about their wealth and all the resources they have to them. so our focus here is to build a

concept that hopefully we can use to try to measure adequately from a wealth side of things or asset side of things along with income side of things. one last plug to the scf, the 2016 results will be available at the end of september. >> thank you very much, kevin. our second discussion today will be bruce meyer, university of chicago. >> thank you. i am very happy to be here. please refer to this handout. i was a late substitution to discuss two 85-page papers. you can think of me as the closer if you want. so these papers are wonderful. the topic is especially important given the magnitude of expenditures on social security, given that the modern government is a pension fund that also happens to have an army.

i like papers also because they hit on the fact that the decline in poverty over the last 50 years has been one of the major anti-poverty achievements in the u.s., and the papers are -- the bee and mitchell paper in particular indicate that that decline in poverty has been even greater than indicated by official statistics. i also like the papers because they support much of what i have been arguing for the last 15 years, they fill in large gaps that i haven't been able to fill in. i have been trying to get the social security administration to do a version of this -- of the bee and mitchell paper for about 15 years.

every few years i submit a proposal, which gets rejected. and on the other hand i don't leak the fact that i'm now in the age of the study group here, my 20-something kids gleefully reported that not once but twice i got the senior discount on a recent trip. so let me quickly summarize first the bee and mitchell paper. they find that about half of pension recipients in our main source of income data for official statistics do not report the pension income they receive. the office then replaced the self-reports of pension and reports of income from some other transfer programs with administrative values, from tax and ssa data.

these corrections of the data lead to many remarkable findings. first, the poverty rate of those 65 and older is much lower. second, the poverty rate falls more over time. though that's not quite how bee and mitchell sell it. if you look at the figures, the divergence between the survey and the administrative numbers get larger over most of the time and then just narrow a bit at the very end. and one should also emphasize that that's a divergence or pretty much constancy in amount on a much smaller base over time. so that, in percentage terms, the reduction is much greater over time. median incomes of those 65 and older are much higher than we thought when we bring in the

administrative data, and median incomes rise much more over time than the survey data alone indicate. the fall in income at retirement is small or non-existent and interesting results. now, to put these results in context, i want to emphasize that all u.s. surveys are seeing deterioration in their quality. so it's not just pension income that's under reported. half of s.n.a.p. isn't reported, and half of cash welfare isn't reported. other work that replaces error ridden survey reports of certain income components with administrative measures sees similarly large effects. i have emphasized in my work the value of looking at consumption data. if you look at consumption-based measures of poverty that look at what people are actually able to spend on -- in terms of their

housing, food, transportation, and other goods and services, and you find much in the way of similarities between the patterns and the consumption data and those in the corrected income data. the corrected being the ones where the administrative data had been substituted for the error ridden survey reports. with jim sullivan, we have argued that, looking at consumption data is conceptually a better way of looking at well-being than reported income. and we conjectured that -- and i'll quote our brookings paper from 2012, the two most plausible explanations for the difference between the changes in income and consumption poverty are measurement error and saving or dis-saving.

there is considerable evidence that the changes in measurement error are important for families with few resources, first transfer income, which is particularly relevant for these families is significantly under reported and surveys and the extent that the under-reporting has grown over time. so there is a lot of similarity with the findings in the papers here. if you look at the figure on the fourth page of the handout, there were reported consumption poverty measures for men and women over the past 30 years, and you will see that, when you look at the open circles which are for men or the open triangles, which are for women, those numbers start to diverge and show lower poverty rates,

sometime around 1990 or a little later. and the consumption measures show a much greater decline in poverty over time for both men and women. again, consistent with what bee and mitchell are finding. the next slide there shows a version of this same issue applied not to the aged but to the overall population. and here we are using data, again, from the cps, that bee and mitchell used, and we are looking at the effect of under-reporting of public assistance, s.n.a.p. which is renamed food stamps and housing benefits. when you use the administrative values rather than survey reports, you find that the poverty reduction for these

three programs combined is 90% greater than if you just use the survey reports and then it differs quite a bit across the programs and combinations of programs that are listed there. so, when you account for the under-reporting of transfers in the household survey data, the effects of these programs in reducing poverty is much greater. so what are the policy lessons here? the elderly are much better off than they appear. it's important because it may affect where we target future anti-poverty efforts. two weeks from now, the census bureau will release the poverty report for last year. and at the same time, the census bureau also release a newfangled poverty measure called the

supplemental poverty measure that, by some arguments, makes some substantial improvements in how we measure poverty. but it does some things that i think are really mistaken. and it leads to a measure of poverty for the elderly that's much higher than the official rate. and one for children that's lower. and so this first order thing that this newfangled poverty measure that will be released along with the official measure two weeks from now, the first-order thing it does that i think is wrong, and that's underlined by what bee and mitchell showed, that in fact the poverty rate of the elderly is much lower than the official measure, because of the

under-reporting of pension income. now, even without this correction, we should have known that poverty was especially overstated for the elderly. almost half of typical spending for a family is housing and transportation, and 80% of those 65 and older already own a home and a car, so that much of the consumption needs are already taken care of and income is not needed to maintain a standard of living as high a standard of living. moving on to changes in income at retirement. bee and mitchell and brady et al, are they complimentary. they tell the same story of little or no decline at retirement.

the bee and mitchell paper uses a longer history of past earnings, so it's able to get a better measure of what baseline income is, typical income. the brady, et al, paper accounts for taxes which is very important since taxes fall a lot in retirement. i do think this literature often goes a little bit awry in focusing on averages and medians. what we really need to get our hands around is what share of people don't have enough to survive on or to do well with in retirement. so it's not really a question of means or medians, it's more about what share of people have

substantial falls in their standard of living. now, the brady, et al, paper finds that there are 10% of people who have a fall of at least 44% three years out from the year before retirement, and we know from bee and mitchell that there has already been a fall in income relative to long-term income right prior to claiming. so there is a substantial -- it depends on how you look at it. how much weight do you put on what happens with 10, 20, 25 percent, depending on how you measure things. a caveat here is that probably there should be more research looking at other sources of

income for this group because probably it isn't really 10% that have a fall of 44% or more because there are income sources that are not captured in the brady, et al, data. it would be useful to bring in things like s.n.a.p. as well and see -- get a more comprehensive picture of the living circumstances of this group. and also i would argue, for looking at consumption. which i think is just a much better way of capturing the living standards of this group, given the durables, houses, cars, that people already have. so the general lessons here are also that survey data alone on income is misleading and that using administrative data by itself or a combination of

survey and administrative data goes a long way in solving the problems with the survey data alone. those at the bottom have more resources than we think, and current institutions, both public and private, reach more of the needy than it appears. >> thank you very much, bruce. i'm cognizant of time here and i want to make sure i leave enough time for audience questions, so i'm going to jump, i also want pau leave enough time for my questions. this may be more for josh, but you can weigh in, as well. what we have to a certain degree is data points. at a point in time. we know how folks today are doing and that can't tell us where people will be in the future, which is when the policy questions we face. at the same time, there's a perception that you know, since

this golden age of pro additional traditional pensions if you've got pension system, retirement incomes should rise with the same rate of earnings of people preretirement. if you had the same rate of increase, you could say at the mean, the adequacy of retirement benefits would stay about the same. josh, you showed some time-serious data looking at changes in retirement incomes as measured through the irs data over time. do you have some idea of how those changes relate to changes in preretirement earnings over that same period? in other words, do you have a feeling for the trend over the past couple of decades of how retirement incomes changed and relative to the preretirement

earnings that retirement plans try to replace? >> that's a research in progress for us. using the longitudinal earnings data for different cohorts and trying to look at least one or two years into retirement relative to those long-run earning averages. if you go back to the start of our series, you can see that the incidents of having any retirement income is much low er. so based on -- >> you say retirement income, you mean income from private retirement. >> from private retirement plans, yes. and so, combining that with that social security rates used to be, i think you could have an estimate, answering my your exact question, my sense is that living standards are probably more able to be maintained when we look in the early otts from the paper and peters if you

looked, say, at in 1990 at this pattern. that might also help explain some of the discrepancies between our paper and the previous literature on the retirement consumption puzzle. so when they were writing these papers on their expenditure, they were looking at people so that's one issue but that were retiring in the 1980s, and perhaps it was the case then that living standards did change more abruptly at retirement. but given that 61% have some sort of private retirement income now, i would expect that living standards are more able to be maintained now than in the past. >> any thoughts. >> well, i think so the first caveat on this is some of what i say is what i think i know and a lot of what i thought i knew isn't right going in. >> when you look to past research, it's undeniable, people have been getting better off. a large part of that is we have had an expansion of social

security. so some people say it's, you know, fdr never meant it to be the only pension plan and that's fine, but we had an expansion in the '50s and another in the '70s. it's a much different program, it's designed to be a program and that's helped a lot. secondly, a lot of times people impute how people got pension income from working. one of the things that happened before erisa came in '74, not everyone who was working at an employer covered by a plan got any of it. some you have to retire from the plan to get any money at all. if you left the day before normal retirement age, you got nothing. in the cps, the pattern holds, whether it's correct or not, you do actually see an increase in pension income in the '70s and '80s and early '90s, costs off at '90. likely, a lot of those are due to the fact that not only that the people who were covered

actually got something from it because we had vesting rule, ten-year tighten to five years in '86. so there are studies that have used the hrs, and then its precursor, i forgot the name and looked at different cohorts. now, the question is is, we've had a bad economy for many years now. we seem to see a stalling in this and the question is, is that a long-term turn around or if we ever recover, will we continue to see the increase go on. >> thank you very much. >> i'd like to open us now up to questions. from the audience. we have some folks with microphones. if you could speak into the mike because we have cspan here. and give your name and your affiliation. we have henry aaron right here in the front. >> broadly speaking as i understand it, the 1983 amendments somewhat reduced replacement rates over the

period since then, the story told is that there has been a growth really in the amounts of income reported from private pensions of various kinds. now, over the same period the preretirement income has grown markedly less equal in its distribution. so, my question and it's a statement, but a question, really. is is it not the case that what your story tells is that the retirement system is doing a very good job of extending into retirement the growing inequality that is observed in earnings and other income prerequirement? so, that you're replacing

faithfully and i take that as a take away from your two papers. you're replacing faithfully, a steady increasing degree of inequality preretirement. >> yeah, i don't think the retirement system is currently set up as -- you have progressive social security benefit schedule. i think a net lower income for the combination of taxes and payment out, probably get more benefits out of it. but in general, the retirement system is is probably not going to counteract things that are happening before retirement and i think it's designed to try to get replacement so it wouldn't really have that effect. >> yes, sir. >> carl. with the project called the center on capital and social equity. first i want to thank bruce myer aei for letting me know that s.n.a.p. is twice as effective

in reducing poverty than we thought, based on administrative data. i thought i would have had to go to the budget for that. but my question is one of my jobs is to look into the future for the long-term care industry and you know, look at their finances and demands of life. you say you're looking in the past. ten, 20, 30 years from now, even when you observe about the transition from d to c. to what, can you be an actuary just pretend? what can you tell me? is it going to be 25%? more income from this administrative data? or a wider vary -- what can i tell my client based on this transition? >> scf might be the best data source we have on this now. >> i think that's you know, part of what we're sort of collecting sort of the wealth. maybe there's some way to take those resources for people at

the requirement and do this projection out to say how are people going to do if they have this pod of money to spend. one of the things that's problematic, there's issues of internal control for people because as you get fact dc plans and they have to manage this, right and so how do they man nlgage that gives any shocks that happen. there's probably a fair amount of people out there that you could say they'd be better off anew anu anu advertising their wealth and retirement to save them from themselves. that's not going to happen. we're trying get an idea of all the sort of resources people would have at retirement with this sort of expanded measure we're trying to get at, then we can we can look at how to stretch this out across life spans using mortality rates and things like that. it's fraught with a lot of assumptions about what's going to happen. but i think the biggest piece of what you're asking is the shift to d.c. is going to you know,

unfortunately increase the uncertainty on that. >> let me make a few comments. i think first of all, economists are good at analyzing data that exists and telling you what happened. we're less accurate forecasting. but there's a few things one is this, there seems to be an opinion that everyone retired with a gold watch and a pension. the fact of the matter is all types of pensions have certain risks. the traditional final average pay has a lot of risk in that the benefits are terribly back loaded. so, you not only have to work for a company for a long time to get it, you have to retire, leave the company close to normal retirement age because a lot of these are set nominally. and so, the idea that everyone particularly low income people, had a whole lot of money from ubs. someone who's long tenured and retires from an employer is not a definition of a low income person. no matter their job. probably at least middle income.

we don't know what's coming up. there are certainly risks on all sides. there are things that you might think are better, particularly for lower income people to the extent and women who tend to move in and out of the workforce, you build over time, it's not as back loaded and we don't know. the other thing is what's going into these things. i think so certainly, there may have been some windfalls that people were getting benefits that were never funded, but as far as money going into the plans, you're looking at combined with employer contribution, employee contributions, close to 11% of pay going into d.c. plans. i don't think we've ever had that going into a defined benefit plan. so, we don't know what the future is going to be. i'm not going to, i don't have a crystal ball. there's certainly uncertainty, but the assumption that we're going from a perfect world to an

imperfect world, there are two imperfect worlds and it's very uncertain as to what's going to happen. when you look at the data, you look at those near retirement. we haven't had a shift in how many people have it. we're starting, the recession caused problems for younger people, but about 80% of working household haves this income and there has been other research certainly with the hrs whether you look where they try to measure this and you see a shift in wealth from db to dc which you aren't seeing much of a decline overall in the benefits. but it's uncertain. i can't tell you what's going to happen in 20 years. >> thanks. yes. >> my name is emily. my question is is what the puerto rico oversight board doing right now, just another union chrysler bailout? >> is that directed at andy? >> that would be directed at me. my day job is at aei.

my what i thought was going to be a part time job, but it turned out to be more than part time is the with federal government's financial oversight and management board for puerto rico. it's whether this is similar to a chrysler bailout puerto rico is facing significant challenges. we have no resources to put into puerto rico, so even if we were inclined to bail them out, we're not able to. i'll say we are not inclined in the sense that i think there's unanimous feeling on the board that puerto rico needs to stuck structural changes to pensions, a whole variety of areas, to make the island sustainable and economically growing over the future. so the short answer is, no, there will be no bailout. unless you could talk to congress, if they werin clibd to do it, but i'm not sure that's going to happen. >> one quick follow up to that. so, in all of these cases, there's security and unsecured

debt. and with chrysler, and both puerto rico, the unsecureded debt includes that of the pension funds. in the case of puerto rico, the institution which passed through congress twice specifically states that the general obligation debt is the most senior and most secure. in chrysler, the case in which what happened was the 55% of union company stock went to the unsecured debt while the secured debt got crammed down to accept a payment of 23 cents on the dollar. conservatives and free market thinkers such as aei were staunchly opposed to the obama bailout of the uaew. so, my question is how is this not the same and as someone presenting the oversight board that has allowed the government to pay 100% of the pension

funds, how can you side with this? >> i'll be quick. this is off topic and you're going to lose most of the viewers on this. you know nothing about puerto rico. the board has instituted pretty dramatic changes to the pension system. about a 10% overall cut in pension benefits. along with large cuts in health care spending, whole range of other things. so there is nobody on the island and nobody, no stakeholders on the island. be it pensioners, be it citizens, be it creditors not making significant sacrifices. other questions. sure, right there. >> i'm bob by. just a citizen of the u.s. interested in -- >> nobody's justice. >> in the issue of requirement. first, i want to compliment the panel on the presentations that were made. which were understandable to a layperson. nice job. i have a question about the work that may be shortly underway in congress.

to change taxes and other benefits. that may be available. and whether you can talk about the effect of what might happen depending upon political perspective, on the kind of findings that you have right now, how that might affect -- >> are you speaking about changes say shifting 401(k)'s or pension to a roth basis? where taxes are paid up front? it wasn't clear to me what tax changes? >> i think the arrangement o potential changes. i guess today or sometime shortly, we're about to hear changing, what changes in tax policy will be presented by the administration. i'm talking about different levels of income and changes in the way in which taxes may be

applied to different levels. >> josh, pete, do you have any thoughts on that? >> it's a very broad topic. i certainly am not going to volunteer what's going to happen. i don't want to suggest something that i don't want to happen, that's for sure. i will say, i have a book on the tax treatment of pensions and potential -- a lot of proposals that have happened in the past and how they affect the tax treatment. and i might refer you to that. and i have some short blogs so if you don't want to read a 220-page book, that sort of describe the issues. don't want to volunteer what might happen for something i don't want. >> josh is not looking eager here. >> i'll stick to measurement. >> yes, ma'am. >> hi, i'm cristina fitzpatrick with aarp. >> i have a question for you. there's a lot in here, both papers that are interesting and

challenge a lot of the assumptions a lot of us have. i'm looking at the grounds for the lowest income. in those, you're saying not only at the median is the replacement rate not going down, but it's actually going up. intuitively, how do we understand that? >> we also show that the increases primarily among those who continue to work after claiming social security. for those no longer working, it's a slight increase. but it's not as big. economists have a danger, we like reading these tables and form stories around them. so i don't want to go too far out on a limb. from looking at the data, one thing that seems to be occurring that conditional and continuing to work. those in the lowest quintile, '99 income, don't have a very large reduction in earnings. whereas the rest of the higher parts of the income distribution, be even those who continue to work, you see a decline in their at least mean earnings and median earnings.

so what it suggests to me is that some at the bottom who are low income may be using this as a wage supplement. maybe they were liquidity constrained. a boost to resources and they're using it to increase their income. maybe they could have used it before they claimed social security. we don't have such plans in place. so i don't want to read too much into a table of numbers and means and medians and incidents. but it looks like those at the bottom are behaving differently. if they continue to work, they're making pretty much what they're making before then adding some additional income. [ inaudible question ] >> there is no decline. >> yes, at the medium. >> so, how is that they're they're not working and just getting social security. >> a lot of them have other sources. so, what we showed about 25% in

the bottom 20% are only relying on social security. you also have to know that the social security system has a very progressive benefit structure. so for a lower earner, you are actually replacing over 100% of your average lifetime earnings, inflation-adjusted. not wage-adjusted, but inflation adjusted. so there's very high replacement rates at the bottom. getting higher rates of social security and also access to -- there are some that have pensions. and they have other income. >> sorry, one more question. is all this inflation adjustment -- >> everything's inflation adjusteded. well, replacement rates are inflation adjusted. we're talking about inflation adjusted income in t plus one prior to income before retirement. >> over the period, you're looking at a wage adjustment wouldn't make much of a difference. >> what i was talking about is the normally, when people are at the bottom of the formula, 90%

of age adjusted lifetime earn, but turns out that at least historically, we hope it happen miss the future, wages have grown faster than inflation. so as a percent of what they actually -- actually earned in real terms, you're actually replacing a much higher percentage. >> just to add one thing. pete's paper does it great. taxes are in there too. you have to understand that the payroll tax going away for people not working is going to be huge. and the fact that social security is not going to be taxed. so they're taxed -- they're sort of net of tax bill goes down, and that, i think, helps also with it. >> i'm glad you read it more recently than i have because yes, i forgot. >> just think if you had somebody in the 90 percent replacement factor, at the very low end, and but they're losing the social security, medicare, payroll tax, 7.65%, if they're not paying taxes in retirement, you're getting close to 100% right there. sir, yes, in the middle. >> ted goldman. senior pension fellow at the

american academy of actuaries. so there is an actuary in the room. my question is getting back to this growing problem of converting savings into lifetime income. and could it be one explanation for t plus three is that people draw down enough asset to draw down their same standard of living, but if you were to continue your study and look at t-10 and t-20, we may see marked reductions in people's ability to maintain the standard of living. so it's good news if people save enough to last three years, but do we know anything beyond that. >> first of all, although we use t 3 in the paper, part is a trade up. the further we look out, the fewer people we can look at. so we basically said, you know, what happens if we look two years out, three years out, four years out. so we have looked at those. there isn't a huge difference. and actually, we have done

sensitivity analysis and looked further back. those don't seem to change it. the other thing we did do is we followed each cohort in '99 and looked at the percentage that had pension income afterward. now, the problem with that, we looked at other ranges oh than 55 to 61. so people who are 70 in '99. people who are 60 in '99 and so forth. the problem with that, i'm pretty sure the 55 to 61-year-olds working, the tax data is simple. nonfilers aren't an issue. i don't know that for the other group. so i can't tell you who is not filing. so for ten years, you line them up as sort of synthetic cohort, you would use, it's basically -- it goes up and doesn't really go

down. now, again, i'm not sure how representative those other groups are, because i don't have the nonfilers and i think nonfiling goes up as you age, for example. but nothing i can see in the data suggests it's going away. but i can't just from the analysis i've done rule it out. >> you don't know how much wealth they've accumulated. >> the only thing we can look at is i.r.a. wealth from the form 5598. we can't look at dc we get oral db wealth and other things. we just have income flows. >> your thoughts. >> we also have the same problem data ending in 2010 so we can't do our synthetic cohort years in the future. but when we look at the cross section, the 85-plus group did have substantial boosts in income relative to the survey. and, in fact, the poverty rates were only somewhat higher for the 85-plus than for the 65 to

74. so obviously, that's mixing cohorts and ages. but that evidence there is consistent with our relatively high ability to maintain some absolute living standard. >> the questions you might want to look at work, consumption, data, something like mike and suzanne from rand who use the hrs cam survey. there is some work with scf. the /* if you look at older retiree, those are folks running out of money, you would expect they'd cut back substantially on discretionary expenditures. gifts to people, charitable contributions, things like that. you don't really see it. but i think the hurd letter which follows individual households over time doesn't find that many people are running out of money. other questions. sure. >> my name is marie, i'm with the economic policy institute. i want to congratulate a very

interesting panel. i really appreciated -- those papers are very interesting. my question is well first of all, quick question for joshua mitchell. you mentioned you're looking at the post 2014 redesign of cps and it's ambiguous whether you're going to do the similar exercise or just talking about how to fix the questions to better capture the retirement income. and i would just -- for those of us who are sort of stuck with using the cps, and want to make corrections for the cps data post 2014, i would very much appreciate if you would repeat the exercise for 2014 and afterwards. the other question i have for both you and kevin moore. i know this is very speculative. the most surprising thing about the paper is how much the db income is under reported. dc income, we knew was underreported. and for obvious reasons. but db income, i mean, kevin

moore also mentioned that it was surprising. if you had any idea why ss to underreported given it's a steady source of income. >> so, we've looked into this in terms of checking if these db withdrawals are one time and they seem to be steady payments. as you say. so it's a bit of a puzzle. we've tried to explain in terms of demographic groups and we don't find much there. i think standard reporting, fatigue, when questions get asked, people may be ready to end the interview. also some notions about what is income may -- people may view this as sort of i earned this during my working life, so they're just not viewing it the way we would, as income. that's another potential issue.

then lastly, i know that fr some other surveys, there's this finding that people tend to report income better when they turn around and spend that income. so for example, if you just get a distribution and it goes into your savings account it could be db or dc. where you have a required minimum distribution. it just goes from one account to another. you might forget to report that as income for the survey. because you didn't, like, spend it that year or that period. >> so the -- one of the things i knew going into this is that we covered db pension income well, but not dc pension income. and i knew all of the explanations for that. i had never opinion satisfied with that explanation and someone who does a survey can explain it better to me. i've never been surveyed, but if you just read the survey questions, the first question literally says do you receive any payments other than social security and part of the things including profit sharing plans.

many may not know this, but 90% are 401(k) plans and vice versa, so the regular payment item isn't was never brought up until the second question which was what is the source of this. and then regular was an option. we've done comparisons of the cross sectional data. the universe, the cps, writing it up. josh's paper came out, maybe some of the things won't be as shocking. but we find many of the same things. and, you know, it can't just be the dc that it's underreporting. it's just not big enough. regular payments and what josh's paper shows is that back in 1990 when there weren't a lot of dcs, we were underreporting this. we were measuring this badly and i don't know if we know why. let me just put a little caveat on, trying to separate this into db and dc. i would really like to separate this in the db and dc.

it's really hard to do with the tax data. i think they have an interesting way to do it. however, i think since there are many things, i don't want to leap to a new thing that i know that really isn't true. so i think some humility is needed that we don't really know exactly where this is coming from. and we can try different ways to get at it. but until it's important for the tax authority to track whether it's a db or dc as much as policymakers want it, it's going to be hard to figure this out. and i think i like going to specific plans, and that -- you know, we can try to narrow it down and identify it, i think we can do some work and identify it. but just a little caution on what's db and what's dc. i don't think we really know. and we don't even know what's a nonqualified annuity payment. >> we're going to have to call time now and i want to thank all of you for coming today and of course, i want to thank all of our speakers who did such a fantastic job. i'm hoping our speakers and discussers will be willing to stay around for a few minutes

afterwards and to answer any questions we didn't get to during the event. thank you all for coming to aie today. thank you. [ applause ] this afternoon, supreme court justice, ruth bader ginsburg speaks with first-year georgetown university law students. our live coverage is scheduled to start at 4:00 p.m. eastern on our companion network, c-span. online at c-span.org, or on the free c-span radio app. i've been on the other end of a phone call from my team asking for my help, because we had received a call from the department of homeland security telling us that a 7-year-old girl was being sexually abused, and that content was being spread around the dark web. and she had been abused and they had watched her for three years. and they could not find the perpetrator. asking us for help. we were the last line of defense.

an actor and his foundation were the potential last line of defense. >> for the past 30 years, the video library is your free resource for politics, congress and washington public affairs. so whether it happened 30 years ago or 30 minutes ago, find it in c-span's video library at c-span.org. c-span, where history unfolds daily. c-span, where history unfolds daily. in 1979, c-span was created as a public service by america's cable television companies. and is brought to you today by your cable or satellite provider. now, federal and local veterans' affairs officials and health care providers discuss better access and specialized care for veterans with spinal cord injuries and